{
  "id": 154785,
  "title": "Adversarial Validation on Pixel Data",
  "url": "/competitions/siim-isic-melanoma-classification/discussion/154785",
  "author_name": "",
  "post_date": "2020-05-29T20:08:19.330564200Z",
  "votes": 6,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I have run an adversarial validation kernel on the resized 32x32 pixel image data. So far I am getting adversarial AUC of 0.65, which is non-trivial. You can find the notebook here:</p>\n\n<p><a href=\"https://www.kaggle.com/tunguz/adversarial-melanoma/\">https://www.kaggle.com/tunguz/adversarial-melanoma/</a></p>",
  "messages": [
    {
      "id": "866943",
      "postDate": "05/29/2020 20:08:19",
      "content": "<p>I have run an adversarial validation kernel on the resized 32x32 pixel image data. So far I am getting adversarial AUC of 0.65, which is non-trivial. You can find the notebook here:</p>\n\n<p><a href=\"https://www.kaggle.com/tunguz/adversarial-melanoma/\">https://www.kaggle.com/tunguz/adversarial-melanoma/</a></p>",
      "rawMarkdown": "I have run an adversarial validation kernel on the resized 32x32 pixel image data. So far I am getting adversarial AUC of 0.65, which is non-trivial. You can find the notebook here:\n\nhttps://www.kaggle.com/tunguz/adversarial-melanoma/",
      "votes": null
    },
    {
      "id": "866978",
      "postDate": "05/29/2020 21:00:39",
      "content": "<p>I didn't know of this technique thanks. How do you usually go from here to build your validation ? Use that model prediction on train only and pick your val set in elements predicted to be test set ?</p>",
      "rawMarkdown": "I didn't know of this technique thanks. How do you usually go from here to build your validation ? Use that model prediction on train only and pick your val set in elements predicted to be test set ?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 866978,
      "author_name": "arroqc",
      "author_url": "",
      "post_date": "05/29/2020 21:00:39",
      "content": "<p>I didn't know of this technique thanks. How do you usually go from here to build your validation ? Use that model prediction on train only and pick your val set in elements predicted to be test set ?</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "866943": "I have run an adversarial validation kernel on the resized 32x32 pixel image data. So far I am getting adversarial AUC of 0.65, which is non-trivial. You can find the notebook here:\n\nhttps://www.kaggle.com/tunguz/adversarial-melanoma/",
    "866978": "I didn't know of this technique thanks. How do you usually go from here to build your validation ? Use that model prediction on train only and pick your val set in elements predicted to be test set ?"
  },
  "source": "meta"
}